OpenAI Responses API

The Responses API uses POST /v1/responses. If your code already calls responses.create from the OpenAI SDK, swap the Base URL and API key for your Tokeness configuration.

Configuration

Base URL: https://n.tokeness.dev/v1
API Key: your Tokeness API key
Model: model name copied from the Tokeness model marketplace

Verification order

1. First send a cURL request to https://n.tokeness.dev/v1/responses.
2. Confirm the request appears in the Tokeness usage logs.
3. Then use the same key, model name, and Base URL in your SDK.

Node.js

Install the SDK:

npm install openai

Create test-tokeness-responses.mjs:

import OpenAI from 'openai'

const client = new OpenAI({
  apiKey: 'YOUR_TOKENESS_API_KEY',
  baseURL: 'https://n.tokeness.dev/v1'
})

const response = await client.responses.create({
  model: 'YOUR_MODEL_NAME',
  input: 'Introduce Tokeness in one sentence.'
})

console.log(response.output_text)

Run:

node test-tokeness-responses.mjs

Python

Install the SDK:

pip install openai

Create test_tokeness_responses.py:

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_TOKENESS_API_KEY",
    base_url="https://n.tokeness.dev/v1",
)

response = client.responses.create(
    model="YOUR_MODEL_NAME",
    input="Introduce Tokeness in one sentence.",
)

print(response.output_text)

cURL

curl https://n.tokeness.dev/v1/responses \
  -H "Authorization: Bearer $TOKENESS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "YOUR_MODEL_NAME",
    "input": "Introduce Tokeness in one sentence."
  }'

Message array format

When you need to distinguish roles, pass input as an array:

{
  "model": "YOUR_MODEL_NAME",
  "input": [
    {
      "role": "developer",
      "content": "Keep your answer short."
    },
    {
      "role": "user",
      "content": "What is the Base URL for Tokeness?"
    }
  ]
}

Troubleshooting

Symptom	Check
404	Is the request path /v1/responses?
401	Is the Tokeness API key complete? Are you using Authorization: Bearer?
Model not supported	Switch to a model from the model marketplace that supports Responses or OpenAI-compatible calls
No output	Print the full response object first to inspect the returned structure and error fields